An online audio spectrogram for any sound file
The Audio Spectrogram turns a recording into a picture of its frequencies over time. Time runs from left to right, pitch from bottom to top, and the colour of each point shows how loud that frequency is at that moment. Open an MP3, WAV, M4A, OGG or FLAC file, zoom into any part, hover to read the exact time, frequency and level, and download the image as a PNG.
Producers and sound engineers use spectrograms to find hum, hiss and clicks before cleaning a recording; audiophiles use them to check whether a “lossless” download was really made from an MP3; birders, linguists and teachers use them to see calls, speech formants and harmonics. It works on long files too — the audio is decoded on your device and nothing is uploaded.
How to use the Audio Spectrogram
- Drop an audio file onto the box or click to choose one. The spectrogram of the whole file appears once it is decoded.
- Pick a Frequency scale: linear to check the high end, logarithmic or mel to see music and speech the way the ear hears them.
- Adjust Window size, Top level and Dynamic range until the detail you want stands out, and choose a colour map.
- Use Zoom in or drag sideways to move through the file; hover (or tap) for the time, frequency, nearest note and level.
- Press play to follow the playhead, then Download PNG — as shown, or at 1920 × 1080 or 3840 × 2160.
Audio Spectrogram features
Real STFT, your settings
A short-time Fourier transform with window sizes from 512 to 16,384 samples, 50–87.5% overlap and Hann, Hamming or Blackman–Harris windows.
Three frequency scales
Linear, logarithmic (20 Hz up) and mel, with a selectable top frequency from 1 kHz to the full range.
Accurate readout
Levels are in dBFS (a full-scale sine reads 0 dB), and the cursor shows time to the millisecond, frequency, the nearest musical note and level.
Zoom without losing peaks
When zoomed out, the frames inside each pixel column are averaged and each pixel row keeps its loudest frequency bin, so short tones and narrow lines do not disappear.
Lossy-source hint
The long-term spectrum is measured to report how high the content really reaches and whether it ends in the sharp shelf typical of MP3 or AAC.
Perceptual colour maps
Magma, Inferno and Viridis (from matplotlib) brighten steadily with level, plus a classic blue–red–yellow map and greyscale for print.
How to read a spectrogram
- Harmonics: a sung or played note is a stack of evenly spaced horizontal lines. The lowest is the fundamental (the pitch you hear); the others sit at 2×, 3×, 4× its frequency and give the instrument its tone.
- Noise: hiss, wind and breath are a broad haze spread over many frequencies rather than lines. Drum hits and clicks are thin vertical streaks — energy at all frequencies for an instant.
- Hum: a steady line at 50 Hz (mains power in India, the UK and Europe) or 60 Hz (the US), often with harmonics at 100/150 Hz or 120/180 Hz, means electrical interference or a ground loop.
- Speech: vowels show as several bright bands (formants) that move as the mouth changes shape; “s” and “sh” sounds are high-frequency noise.
- Lossy encoding: MP3, AAC and Vorbis encoders discard content above a cut-off that depends on the bitrate — typically around 16–17 kHz at 128 kbit/s and 19–20 kHz at 256–320 kbit/s, depending on the encoder. A flat “ceiling” across the whole song at one of these heights in a FLAC or WAV file means it was converted from a lossy file.
Choosing the window size
A longer window separates frequencies more finely but blurs timing; a shorter one shows timing sharply but smears pitch. Values below are for 44.1 kHz audio.
| Window | Duration | Frequency spacing | Good for |
|---|---|---|---|
| 512 | 11.6 ms | 86 Hz | Drum hits, clicks, precise timing |
| 1024 | 23.2 ms | 43 Hz | Speech, fast passages |
| 2048 | 46.4 ms | 21.5 Hz | General music (default) |
| 4096 | 92.9 ms | 10.8 Hz | Bass notes, harmonics |
| 8192–16384 | 186–372 ms | 5.4–2.7 Hz | Hum, steady tones, tuning |
Related tools
Generate test signals with the Tone Generator or Noise Generator and look at them here, make a clean waveform image with the Audio Waveform Generator, or find a song’s key with the Key Finder.
Audio Spectrogram FAQ
What is an audio spectrogram?
It is a graph of sound with time on the horizontal axis, frequency on the vertical axis and loudness shown as colour. It reveals things a waveform cannot, such as pitch, harmonics, hum and missing high frequencies.
How can I tell if a FLAC file is fake?
Open it here with the linear scale and full range. A genuine CD-quality recording usually has content fading out towards 20–22 kHz; a file converted from MP3 or AAC shows a hard, flat cut-off, often near 16 kHz, and the tool’s bandwidth note will say so.
Which scale should I use?
Linear spreads frequencies evenly and is best for checking the high end. Logarithmic and mel give more room to low and mid frequencies, matching how we hear pitch, so melodies, bass lines and speech are easier to see.
What do the dB values mean?
Levels are in decibels relative to digital full scale: a pure sine at maximum level reads about 0 dB, and every 20 dB lower is ten times smaller in amplitude. Lower the top level or the dynamic range to hide quiet noise.
Is my audio uploaded?
No. Decoding and the Fourier analysis run inside your browser, so private recordings stay on your device.
Why does a long file render at a lower sample rate?
To keep memory use reasonable, very long files (over about 30 million samples) are analysed at half or a third of their sample rate, which hides only the highest frequencies. The file details show when this happens.